Week of May 4, 2026
May the fourth be with you.
Last weekend, as the lights dimmed at Baltimore's Hippodrome, I was repeatedly tapping my phone, prompting Claude to code on my laptop back at home. I was there to see Kimberly Akimbo, a sardonic musical about a girl with progeria, a terminal genetic disease, and what it means to live well when time is short. Somewhere in act one, as the cast sang about the play's central tension — how to balance getting the most out of your time (especially when it's short) with making that time meaningful — I turned off my phone.
The last few weeks of vibe coding have been intoxicating. I've experienced dopamine surges that make doom-scrolling feel tedious. The rate-limiting step in bringing my ideas to fruition has shifted from how fast I can type to how many hours I can spend talking with Claude. My sanity will depend on rejecting the temptation to explore every rabbit hole and (seemingly good) idea.
I'm glad I paid attention to the play. It included the best musical number I have ever heard about scurvy (and yes, there are two other songs about scurvy, one by Pink from SpongeBob and another by The Gregory Brothers).
The Google Notebook LM AI-generated podcast version of this week’s newsletter.
Science and Technology Trends
A variety of scientists, including some from Novo Nordisk (maker of Ozempic), released data on the first quintuple-molecule weight-loss drug: a combination of GLP-1, GIP, and a triple PPAR nuclear receptor agonist. The data imply that this new combination molecule is more effective across multiple domains (weight loss, fibrosis, and appetite control) than semaglutide and tirzepatide (GLP-1 and GLP-1 + GIP agonists). While this study examines the use of this new quintuple-molecule pharmaceutical across various strains of mice (including obese mice), it appears we have unlocked and are evolving the keys to the biologic mechanisms of appetite, dopamine-based reward systems, and body-mass regulation. At this rate, we will have goal weights in the negative integers.
Article:
AI-Summary:
In the last few weeks, I've seen numerous articles about intracellular structures (in brain cells) called microtubules and their role in consciousness. Some of the articles and comments fall into the realm of “astral conspiracy” [i.e., microtubules are the brain's connection to a greater quantum-entangled consciousness; see the Penrose-Hameroff "Orch-OR" theory], while others are far more grounded. Last week, I found an article describing how a group of Wellesley neuroscience students discovered that rats given the drug Epothilone B to "stabilize" neuronal microtubules responded significantly more slowly to anesthetic agents, suggesting that anesthesia disrupts some aspect of microtubules. Whether microtubules are multi-dimensional quantum antennae or simply one of several molecular handles through which anesthesia takes hold is unclear, but the convergence of astral-conspiracy framing and bench-top physiology is itself worth paying attention to. Sort of like my late mother's take on why she bought the National Enquirer: "They seem to know something about what's going on, so I read it to know what to pay attention to."
Press Release-like article on the research:https://www.wellesley.edu/news/wellesley-teams-new-research-on-anesthesia-unlocks-important-clues-about-the-nature-of-consciousness
AI-Summary of the paper referenced:
I had Claude summarize some of the wide-ranging microtubule data (or lack thereof). Long story short - all we know at the moment is that microtubules are, somehow, related to consciousness.
The best part of the latest research on yawning is reading the linked paper’s methods section to learn that 22 healthy adults lay in an MRI scanner with real-time flow imaging while watching videos of yawning humans and animals to trigger a “contagious yawn” (where does one get videos of yawning animals and people?). The research indicates that yawning may be a physiologic means of coordinating cerebral blood flow with cerebrospinal fluid flow, but it still does not answer why yawning is necessary and common across so many species. (I bet you feel the urge to yawn after reading about yawning.)
Article:
AI-Summary:
Anti-Anti-Science
Keeping it light this week: Sometimes I find myself stuck in a rabbit hole when my plane’s wifi drops, and I last loaded a delightful article from 1945 on the possibility that one day, “an individual will store all his books, records, and communications, [in a way] which is mechanized so that it may be consulted with exceeding speed and flexibility. It is an enlarged intimate supplement to his memory.” I suggest reading about the life of the author, Vannevar Bush. When you invest in competent people and science, science fiction turns into science fact.
AI-assisted review of this article and Vannevar Bush:
AI Impact
British researchers demonstrated in a large-scale trial that people who believe they are speaking to a clinical AI chatbot may not provide as detailed symptom descriptions, thereby limiting the chatbot's ability to offer sound advice. While the study used hypothetical vignettes, participants were told to act as though they had the flu or an "unusual headache.” The study included 500 participants, who were randomized to the chatbot vs. human respondent cohorts. The clinical significance of these data is not elucidated by the study. However, this study is another piece of evidence that humans dealing with AI bring a unique set of cognitive biases to the discussion/interaction — humans assume the AI knows more than it does, underappreciate data quality, and underrate the need for robust context.
Article:
AI Analysis:
A loyal reader and I both found this article this week: Harvard and Stanford researchers used OpenAI's GPT-4 o1-preview reasoning model across six clinical-reasoning experiments. The LLM outperformed both prior models (GPT-4) and physicians (including two attending internists evaluating real, previously unprocessed emergency-department charts). Of note, the LLM reviewed text-only internal-medicine and ER cases from an EHR. And there is no evidence (yet) that reading a chart well translates to safer patient outcomes when an AI is actually deployed in a workflow. However, it is clear that this kind of data reinforces why humans tend to see AI as all-knowing and are quick to accept an answer from a chatbot, even if it lacks appropriate context or is accessing poor-quality data.
AI Analysis:
The accompanying editorial is worth reading, as it explores the implications of tools that, on the surface, are better at diagnosis than humans.
AI Analysis:
Things I learned this week
Elite Event Robotics, a company that rents high-end humanoid robotics for entertainment and other events, found that Bebop, its humanoid robot (which was booked to fly with its team on a recent Southwest flight out of San Diego), ran into FAA and Southwest regulations. Apparently, Bebop's large battery raised concerns that it exceeded limits and needed to be disconnected and shipped separately. I had not contemplated the interaction between FAA/airline policies and robotics. I wonder how TSA validates a robot's operating system to ensure it won’t go rogue mid-flight? (or, to paraphrase Philip K Dick, Do Androids Dream of Airline Points and Tier Status?)
Low-key alarmist medical headline of the week: “Rat Lungworm Is Now in My State. Is It in Yours Too? The global, brain-invading parasite continues to spread.” I’ll let you read the article, but the words “Rat,” “Lungworm,” “Now,” and “My State” make me uneasy.
AI Summary:
Barilla has a Spotify playlist for each of their pasta types: “[The] nine to 11 minutes required for perfectly cooked Barilla spaghetti, penne, fusilli, and linguine are paired with songs precisely timed to start and end in the exact number of minutes it takes for perfect al dente preparation.” I found the 11-minute Penne Rigate No. 73 to be particularly uplifting and surprisingly R-rated Old Thing Back (it is a Notorious B.I.G. song). Someone should seek funding to see if the lyrics affect pasta stiffness.
Article:
Barilla Playlists by pasta type
AI art of the week
A visual mashup of topics from the newsletter, and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
A visual mashup of topics from the newsletter, and an exercise to see how various LLMs interpret the prompt. I use an LLM to summarize the newsletter, suggest prompts, and generate images with different LLMs.
This week's prompt asks each model to render a US Patent Office drawing for the Multi-Purpose Existential Apparatus (Patent No. WAiR-0504-2026) — a single absurdly over-engineered humanoid invention that integrates every theme from the issue into a deadpan blueprint: a cranial microtubule resonance array, a forehead-mounted yawn induction viewport, a five-canister GLP-1/GIP/triple-PPAR injection manifold reading "Goal Weight: −12 kg," a Vannevar Bush memex card-catalog drawer in the abdomen, a pasta-timing carburetor calibrated to 11 minutes, a rat lungworm containment vessel, a scurvy prevention compartment holding a single lime per Kimberly Akimbo Act I, and a Southwest Airlines boarding pass noting that the battery has been disconnected per FAA regulation. The full prompt and per-model notes are linked here: https://claude.ai/public/artifacts/d724fa4e-b3e9-40fb-8871-659eadc00eb4.
Clean, your hands must be; sharp, your mind must stay,
-Adam
Comments
Post a Comment